GShard: Scaling Giant Models with Conditional Computation and Automatic Sharding

Neural network scaling has been critical for improving the model quality in many real-world machine learning applications with vast amounts of training data and compute. Although this trend of scaling is affirmed to be a sure-fire approach for better model quality, there are challenges on the path s...

Full description

Saved in:
Bibliographic Details
Published inarXiv.org
Main Authors Lepikhin, Dmitry, Lee, HyoukJoong, Xu, Yuanzhong, Chen, Dehao, Firat, Orhan, Huang, Yanping, Krikun, Maxim, Shazeer, Noam, Chen, Zhifeng
Format Paper
LanguageEnglish
Published Ithaca Cornell University Library, arXiv.org 30.06.2020
Subjects
Accelerators
Annotations
Machine learning
Machine translation
Neural networks
Parallel processing
Online AccessGet full text

Cover

More Information
Summary:Neural network scaling has been critical for improving the model quality in many real-world machine learning applications with vast amounts of training data and compute. Although this trend of scaling is affirmed to be a sure-fire approach for better model quality, there are challenges on the path such as the computation cost, ease of programming, and efficient implementation on parallel devices. GShard is a module composed of a set of lightweight annotation APIs and an extension to the XLA compiler. It provides an elegant way to express a wide range of parallel computation patterns with minimal changes to the existing model code. GShard enabled us to scale up multilingual neural machine translation Transformer model with Sparsely-Gated Mixture-of-Experts beyond 600 billion parameters using automatic sharding. We demonstrate that such a giant model can efficiently be trained on 2048 TPU v3 accelerators in 4 days to achieve far superior quality for translation from 100 languages to English compared to the prior art.
ISSN:2331-8422